EMSS 2011 Proceeding

An Asynchronous Parallel Hybrid Optimization Approach To Simulation-Based Mixed-Integer Nonlinear Problems

Authors:   Kathleen Fowler, Timothy Kopp, Jacob Orsini, Josh Griffin, Genetha Gray

Abstract

To address simulation-based mixed-integer problems, a hybrid algorithm was recently proposed that combines the global search strengths and the natural capability of a genetic algorithm to handle integer variables with a local search on the real variables using an implementation of the generating set search method. Since optimization is guided only by function values, the hybrid is designed to run asynchronously on a parallel platform. The algorithm has already been shown to perform well on a variety of test problems, and this work is a first step in understanding how the parallelism and local search components influence the search phase of the algorithm. We show that the hybridization can improve the capabilities of the genetic algorithm by using less function evaluations to locate the solution and provide a speed-up analysis on a standard mixed-integer test problem with equality and inequality constraints.

I3M  Scientific Sponsors

I3M  Industrial Sponsors

I3M  Media Sponsors